Nearest Neighbor-Based Strategy to Optimize Multi-View Triplet Network for Classification of Small-Sample Medical Imaging Data.

Multi-view classification with limited sample size and data augmentation is a very common machine learning (ML) problem in medicine. With limited data, a triplet network approach for two-stage representation learning has been proposed. However, effective training and verifying the features from the representation network for their suitability in subsequent classifiers are still unsolved problems. Although typical distance-based metrics for the training capture the overall class separability of the features, the performance according to these metrics does not always lead to an optimal classification. Consequently, an exhaustive tuning with all feature-classifier combinations is required to search for the best end result. To overcome this challenge, we developed a novel nearest-neighbor (NN) validation strategy based on the triplet metric. This strategy is supported by a theoretical foundation to provide the best selection of the features with a lower bound of the highest end performance. The proposed strategy is a transparent approach to identify whether to improve the features or the classifier. This avoids the need for repeated tuning. Our evaluations on real-world medical imaging tasks (i.e., radiation therapy delivery error prediction and sarcoma survival prediction) show that our strategy is superior to other common deep representation learning baselines [i.e., autoencoder (AE) and softmax]. The strategy addresses the issue of feature's interpretability which enables more holistic feature creation such that the medical experts can focus on specifying relevant data as opposed to tedious feature engineering.

Reconstructing sibling relationships in wild populations.

Reconstruction of sibling relationships from genetic data is an important component of many biological applications. In particular, the growing application of molecular markers (microsatellites) to study wild populations of plant and animals has created the need for new computational methods of establishing pedigree relationships, such as sibgroups, among individuals in these populations. Most current methods for sibship reconstruction from microsatellite data use statistical and heuristic techniques that rely on a priori knowledge about various parameter distributions. Moreover, these methods are designed for data with large number of sampled loci and small family groups, both of which typically do not hold for wild populations. We present a deterministic technique that parsimoniously reconstructs sibling groups using only Mendelian laws of inheritance. We validate our approach using both simulated and real biological data and compare it to other methods. Our method is highly accurate on real data and compares favorably with other methods on simulated data with few loci and large family groups. It is the only method that does not rely on a priori knowledge about the population under study. Thus, our method is particularly appropriate for reconstructing sibling groups in wild populations.

Predictability analysis for an automated seizure prediction algorithm.

Epileptic seizures of mesial temporal origin are preceded by changes in signal properties detectable in the intracranial EEG. A series of computer algorithms designed to detect the changes in spatiotemporal dynamics of the EEG signals and to warn of impending seizures have been developed. In this study, we evaluated the performance of a novel adaptive threshold seizure warning algorithm (ATSWA), which detects the convergence in Short-Term Maximum Lyapunov Exponent (STLmax) values among critical intracranial EEG electrode sites, as a function of different seizure warning horizons (SWHs). The ATSWA algorithm was compared to two statistical based naïve prediction algorithms (periodic and random) that do not employ EEG information. For comparison purposes, three performance indices "area above ROC curve" (AAC), "predictability power" (PP) and "fraction of time under false warnings" (FTF) were defined and the effect of SWHs on these indices was evaluated. The results demonstrate that this EEG based seizure warning method performed significantly better (P < 0.05) than both naïve prediction schemes. Our results also show that the performance indexes are dependent on the length of the SWH. These results suggest that the EEG based analysis has the potential to be a useful tool for seizure warning.

Brain response pattern identification of fMRI data using a particle swarm optimization-based approach.

Many neuroscience studies have been devoted to understand brain neural responses correlating to cognition using functional magnetic resonance imaging (fMRI). In contrast to univariate analysis to identify response patterns, it is shown that multi-voxel pattern analysis (MVPA) of fMRI data becomes a relatively effective approach using machine learning techniques in the recent literature. MVPA can be considered as a multi-objective pattern classification problem with the aim to optimize response patterns, in which informative voxels interacting with each other are selected, achieving high classification accuracy associated with cognitive stimulus conditions. To solve the problem, we propose a feature interaction detection framework, integrating hierarchical heterogeneous particle swarm optimization and support vector machines, for voxel selection in MVPA. In the proposed approach, we first select the most informative voxels and then identify a response pattern based on the connectivity of the selected voxels. The effectiveness of the proposed approach was examined for the Haxby's dataset of object-level representations. The computational results demonstrated higher classification accuracy by the extracted response patterns, compared to state-of-the-art feature selection algorithms, such as forward selection and backward selection.

An integrated feature ranking and selection framework for ADHD characterization.

Today, diagnosis of attention deficit hyperactivity disorder (ADHD) still primarily relies on a series of subjective evaluations that highly rely on a doctor's experiences and intuitions from diagnostic interviews and observed behavior measures. An accurate and objective diagnosis of ADHD is still a challenge and leaves much to be desired. Many children and adults are inappropriately labeled with ADHD conditions, whereas many are left undiagnosed and untreated. Recent advances in neuroimaging studies have enabled us to search for both structural (e.g., cortical thickness, brain volume) and functional (functional connectivity) abnormalities that can potentially be used as new biomarkers of ADHD. However, structural and functional characteristics of neuroimaging data, especially magnetic resonance imaging (MRI), usually generate a large number of features. With a limited sample size, traditional machine learning techniques can be problematic to discover the true characteristic features of ADHD due to the significant issues of overfitting, computational burden, and interpretability of the model. There is an urgent need of efficient approaches to identify meaningful discriminative variables from a higher dimensional feature space when sample size is small compared with the number of features. To tackle this problem, this paper proposes a novel integrated feature ranking and selection framework that utilizes normalized brain cortical thickness features extracted from MRI data to discriminate ADHD subjects against healthy controls. The proposed framework combines information theoretic criteria and the least absolute shrinkage and selection operator (Lasso) method into a two-step feature selection process which is capable of selecting a sparse model while preserving the most informative features. The experimental results showed that the proposed framework generated the highest/comparable ADHD prediction accuracy compared with the state-of-the-art feature selection approaches with minimum number of features in the final model. The selected regions of interest in our model were consistent with recent brain-behavior studies of ADHD development, and thus confirmed the validity of the selected features by the proposed approach.

Adaptive epileptic seizure prediction system.

Current epileptic seizure "prediction" algorithms are generally based on the knowledge of seizure occurring time and analyze the electroencephalogram (EEG) recordings retrospectively. It is then obvious that, although these analyses provide evidence of brain activity changes prior to epileptic seizures, they cannot be applied to develop implantable devices for diagnostic and therapeutic purposes. In this paper, we describe an adaptive procedure to prospectively analyze continuous, long-term EEG recordings when only the occurring time of the first seizure is known. The algorithm is based on the convergence and divergence of short-term maximum Lyapunov exponents (STLmax) among critical electrode sites selected adaptively. A warning of an impending seizure is then issued. Global optimization techniques are applied for selecting the critical groups of electrode sites. The adaptive seizure prediction algorithm (ASPA) was tested in continuous 0.76 to 5.84 days intracranial EEG recordings from a group of five patients with refractory temporal lobe epilepsy. A fixed parameter setting applied to all cases predicted 82% of seizures with a false prediction rate of 0.16/h. Seizure warnings occurred an average of 71.7 min before ictal onset. Similar results were produced by dividing the available EEG recordings into half training and testing portions. Optimizing the parameters for individual patients improved sensitivity (84% overall) and reduced false prediction rate (0.12/h overall). These results indicate that ASPA can be applied to implantable devices for diagnostic and therapeutic purposes.

A gradient-based adaptive learning framework for online seisure prediction.

Most of the current epileptic seizure prediction algorithms require much prior knowledge of a patient's pre-seizure electroencephalogram (EEG) patterns. They are impractical to be applied to a wide range of patients due to a high inter-individual variability of pre-seizure EEG patterns. This paper proposes an adaptive prediction framework, which is capable of accumulating knowledge of pre-seizure EEG patterns by monitoring long-term EEG recordings. The experimental results on five patients indicate that the adaptive prediction framework is effective to improve prediction accuracy over time and thus achieve a personalized seizure predication for each patient.

Combinatorial reconstruction of half-sibling groups from microsatellite data.

While full-sibling group reconstruction from microsatellite data is a well-studied problem, reconstruction of half-sibling groups is much less studied, theoretically challenging, and computationally demanding. In this paper, we present a formulation of the half-sibling reconstruction problem and prove its APX-hardness. We also present exact solutions for this formulation and develop heuristics. Using biological and synthetic datasets we present experimental results and compare them with the leading alternative software COLONY. We show that our results are competitive and allow half-sibling group reconstruction in the presence of polygamy, which is prevalent in nature.

Optimisation and data mining techniques for the screening of epileptic patients.

Identifying abnormalities or anomalies by visual inspection on neurophysiologic signals such as ElectroEncephaloGrams (EEGs), is extremely challenging. We propose a novel Multi-Dimensional Time Series (MDTS) classification technique, called Connectivity Support Vector Machines (C-SVMs) that integrates brain connectivity network with SVMs. To alter noise in EEG data, Independent Component Analysis based on the Unbiased Quasi Newton Method was applied. C-SVM achieved 94.8% accuracy classifying subjects compared to 69.4% accuracy with standard SVMs. It suggests that C-SVM can be a rapid, yet accurate, technique for online differentiation between epileptic and normal subjects. It may solve other classification MDTS problems too.

Feature extraction of linear predictors at spectral bands of interest.

Intra-cranial electroencephalograms (EEG) from two patients diagnosed with epilepsy are sampled at 1 kHz, enabling analysis and feature extraction at frequency bands above the gamma range. This study focuses on the extraction of linear features (including autoregressive, autoregressive-moving average and Fourier coefficients) obtained at both low (below 100 Hz) and high (100-500 Hz) bands of the signal spectrum. Comparisons of the performance of each feature are made based on a binary hypothesis test of statistical distributions from inter-ictal and pre-ictal epochs. Results are obtained from pre-ictal time periods as assessed by an expert epileptologist.

Error tolerant sibship reconstruction in wild populations.

Kinship analysis using genetic data is important for many biological applications, including many in conservation biology. Wide availability of microsatellites has boosted studies in wild populations that rely on the knowledge of kinship, particularly sibling relationships (sibship). While there exist many methods for reconstructing sibling relationships, almost none account for errors and mutations in microsatellite data, which are prevalent and affect the quality of reconstruction. We present an error-tolerant method for reconstructing sibling relationships based on the concept of consensus methods. We test our approach on both real and simulated data, with both pre-existing and introduced errors. Our method is highly accurate on almost all simulations, giving over 90% accuracy in most cases. Ours is the first method designed to tolerate errors while making no assumptions about the population or the sampling.